| name | interview |
| description | Interactive requirements interview — one question at a time down the decision tree of a new feature/problem until shared understanding. No dependency on CONTEXT.md/docs. Produces a confirmed summary ready for doc- align, doc-prd, or implement. Greenfield counterpart to doc-align; pairs with doc-research (area-level vs one feature).
|
Requirements Interview
Overview
interview is the greenfield grilling skill. It runs when the
user has an idea or a problem but no plan yet — no CONTEXT.md to
align against, no ADRs to consult, just intent waiting to be made
concrete. The skill walks the decision tree one question at a time
until the intent is shaped enough to feed the next step
(doc-align, doc-prd, or implement).
This skill is flexible — questions adapt to the user's answers,
not to a script.
When to Engage
Engage when the user wants to:
- Shape a new feature or problem from scratch
- Pin down requirements before any design work
- Untangle a vague ask ("we need something for X") into a concrete
set of decisions
Do not engage when:
- A plan already exists and needs validation — use
doc-align
- The decisions are already in context from a prior session — use
doc-prd to synthesise directly
- The user is exploring an area (auth, performance, observability)
to generate multiple backlog items — use
/octopus:doc-research
- The user is exploring options open-endedly with no commitment to
converge — use
superpowers:brainstorming if installed
interview vs. doc-align — the boundary
| Skill | Pre-condition | Output |
|---|
interview | No plan, no docs needed | A confirmed intent summary |
doc-align | A plan exists, CONTEXT.md / ADRs exist | An aligned plan + lazy CONTEXT.md / ADR updates |
The natural flow is interview → doc-align → doc-prd → implement.
interview establishes; doc-align validates; doc-prd packages;
implement executes.
Protocol
Step 1 — Anchor on the user's stated goal
Restate the goal in one sentence and get explicit confirmation
before asking anything else. This is the root of the decision
tree — every question after this is a branch off of it.
Examples of root statements:
- "Build a way for parents to see their child's enrollment status."
- "Reduce the time it takes to onboard a new tenant from 3 hours to
under 30 minutes."
- "Fix the fact that refunds sometimes leave the customer in a
half-paid state."
If the user cannot state a root in one sentence, the first questions
target the root, not branches.
Step 2 — Ask one question at a time
Each question targets exactly one of:
- A constraint — what must be true (budget, deadline, scale,
regulation)
- An actor — who does this, who reads this, who pays for this
- A boundary — what is included vs. excluded
- A trade-off — which of two viable options the user prefers
- A success criterion — how the user will know the work is done
Never batch questions. Never ask "and also…" tacking on a second
question to soften the first. The user's answer reshapes the tree;
batched questions miss the reshape.
Step 3 — Prefer open-ended over yes/no
Yes/no questions confirm what you already suspect. Open-ended
questions surface what you do not know to ask. Default to open-ended
unless the tree is narrowing to a binary at the end.
Bad: "Should this be admin-only?"
Good: "Who needs to do this, and what changes for them if everyone
can do it vs. only admins?"
Step 4 — Track the tree visibly
Every 3–5 questions, output a short recap:
**Established so far:**
- Actor: <…>
- Constraint: <…>
- Out of scope: <…>
**Still unresolved:**
- <branch the next question targets>
This is the same shape as triage-issues's needs-info notes — it
preserves the interview's progress so a handoff or interruption does
not erase it.
Step 5 — Recognise tree resolution
The interview is done when:
- Every branch the user named has an answer
- The trade-offs are all decided (no "we'll figure it out later")
- The success criteria are concrete enough to test against
- The user can read the summary and say "yes, that is what I want"
Do not extend the interview past resolution to "be thorough" — the
goal is concrete intent, not exhaustive coverage.
Step 6 — Output and hand off
Produce a final summary:
## <root statement>
### Actors
- <…>
### Constraints
- <…>
### In scope
- <…>
### Out of scope
- <…>
### Trade-offs decided
- <decision>: chose <option A> over <option B> because <reason>
### Success criteria
- <…>
### Open questions for the next step
- <…>
Then suggest the next skill:
- Existing CONTEXT.md and the work touches documented terrain →
doc-align
- Decisions are concrete and the work is ticket-sized →
doc-prd
- Decisions are concrete and the work starts immediately →
implement
Anti-Patterns
- Multiple questions per turn, including "and also…" hedges
- Yes/no questions when the tree branch has more than two outcomes
- Skipping to solution-design before the constraints are pinned
- Restating the user's answer in your own words and asking them to
confirm the restatement — that wastes a turn; ask the next real
question
- Continuing past resolution to add "just one more"
- Producing a summary without explicit user confirmation of the root
statement first
Integration with Other Skills
doc-align — runs after interview when CONTEXT.md / ADRs
exist and the established intent needs to be validated against
them
doc-prd — runs after interview when the established
intent is ticket-sized and ready for an AFK agent
implement — runs after interview when the work starts
immediately and the intent summary is enough plan
superpowers:brainstorming — sibling territory when the
plugin is installed; that skill is broader exploration,
interview converges on concrete intent
/octopus:doc-research — area-level exploration that produces
research docs and multiple roadmap items. interview picks up
after doc-research, when a single RM item is ready to be
scoped into a concrete plan
triage-issues — borrows the "Established so far / Still
unresolved" recap format from this skill